Researcher finds 26,400 valid cloud and AI credentials in public Gitee repositories

TLDR

• Ben Zimmermann found 26,400 working credentials in 7.25 million public Gitee repositories; 9,473 were Alibaba Cloud AccessKey pairs.¹²
• He emailed vendors with the most exposed keys but did not contact developers; revocation was unconfirmed at publication.¹²
• Scanners read full commit history, so keys removed later were still counted; each scanner alone found about half.¹²

Independent researcher Ben Zimmermann reports scanning 7,256,531 public repositories on Gitee, which the post describes as China's national open-source code hosting platform, and finding 26,400 unique credentials that authenticated against their issuing vendors' APIs at scan time.¹ The Hacker News reports that most were keys for cloud and AI services, led by 9,473 Alibaba Cloud AccessKey pairs, followed by 3,805 YouTube API keys, 3,170 DeepSeek keys, 1,489 keys for Zhipu's BigModel platform and 1,348 OpenAI keys, with smaller numbers for GitHub, AWS, Google Cloud and Gemini.² Zimmermann also reports one "active organization-admin credential for a major enterprise AI platform"; The Hacker News says he did not name the platform or say what the credential could reach.¹² His post puts Alibaba Cloud exposure at roughly 1 in every 766 Gitee repositories and gives per-scanner Alibaba counts of 9,428 for TruffleHog and 993 for Kingfisher, while the combined 9,473 figure comes from The Hacker News.¹²

He ran TruffleHog and Kingfisher over every repository, and each tried candidate credentials against the issuing service and discarded those that failed.¹ The scanners read each repository's full commit history, so a key a developer later removed could still be found and counted.² The post says the 26,400 total is deduplicated by secret value, that TruffleHog found 12,737 and Kingfisher 14,664, and that only 1,001 were found by both.¹ It adds that 12,242 of the credentials were committed in 2025.¹ The scan took roughly a month of continuous execution, but The Hacker News notes that Zimmermann did not give the dates, so it is unclear how long ago the credentials were tested, and he did not say whether anyone else had used them.¹² The guest post was developed with the Kingfisher team at MongoDB, which publishes it, and the compute cost of $3,097.64 was covered by a $5,000 Modal sponsorship.¹

On disclosure, Zimmermann says he emailed the vendors with the highest exposure volume a full list of verified-live credentials for their service, plus source-repository metadata to help them identify affected customers.¹ Contacting each committer was "not pursued at this scale", he writes, and revocation status for the disclosed credentials had not yet been confirmed at the time of writing.¹ The Hacker News adds that he did not contact the developers who leaked the keys and that anyone who has pushed a key to a public Gitee repository cannot assume it has been revoked; it cites Truffle Security as noting that deleting a key from code or rewriting history does not revoke it.² Most credentials were committed from personal mail services such as qq.com, 163.com and gmail.com, and the post says corporate committer domains, which include Huawei, ByteDance, Google, Microsoft and Oracle, are a floor.¹ It also says most Chinese companies lack bug bounty programmes or accessible security channels, leaving leaked credentials without a clear remediation path.¹

On comparison, Gitee's rate of 3.64 working credentials per 1,000 repositories is higher than the earlier GitLab (3.11) and Bitbucket (2.39) scans.¹² Those scans used TruffleHog alone, and counting only TruffleHog's results puts Gitee at 1.76, so The Hacker News says neither figure shows which platform leaks more.¹² Neither source includes a response from Gitee or the affected vendors.¹²

Why it matters: keys committed to a public repository can keep working until the vendor revokes them, and the researcher says revocation of the disclosed credentials was not yet confirmed when he published.¹²